What are the best ways to ensure interpretability and explainability of decision analysis?

The context within which you are distinguishing between interpretability and explainability is also very important. For example, in analyzing customer behavior you face challenges, especially when dealing with complex data-driven models. You typically have a trade-off between accuracy and Interpretability: Highly accurate models, such as DNN can sacrifice interpretability. Finding the right balance between model accuracy and the ability to interpret or explain decisions becomes crucial, especially when extracting customer-related insights. The communication gap between data scientists and stakeholders due to understanding the intricacies of complex models' limitations when discussed with non-technical stakeholders can be challenging.

One thing to add, Interpretability is often the first step before explainability. Unless a user is clear on multiple interpretations possible on a decision analysis the explanation might be restricted. Further it is often important to add a legend on readability of the model.

Very critical factors which influences the model to be chosen would be 1. Outcome 2. Users 3. Underlying data and it's inter dependency

When dealing with financial data and aiming for interpretability and explainability in decision analysis, the selection should align with the specific characteristics such as the nature of the decision problem, and the needs of stakeholders. Some key considerations: - Data Characteristics: Consider the volume and variety of financial data. Methods, like Bayesian networks, require a larger volume of data, while simpler methods may be more robust with limited data. - Audience expertise: If the audience is from the business side, opt for methods that are more intuitive and require less specialized expertise. Regulatory requirements: Assess whether there are regulatory requirements governing decision-making in the financial domain.

The decision models are consumed by humans which makes them a critical part of the whole process. It is important to understand the underlying business complication with experts and data elements and improve the model.

While Visualisation could help the understanding, it is important to note the same can't be an overkill. The focus should be using visualisation as a tool to convey the explanation and not the other way about. My experience is sometimes visuals take a front seat leaving the objective in total.

Every model contains assumptions and dependencies, someone who is not part of the ecosystem might not be aware of the same. It is important to provide documentation covering these layers including justification, outliers and exceptions.

Documentation provides a clear and detailed explanation of the logic, assumptions, and methodologies used in the decision model. This transparency aids stakeholders, including non-technical ones, in understanding how decisions are reached. It also serves as a common reference point for all stakeholders involved in the decision process. It facilitates effective communication and collaboration among team members, analysts, and decision-makers. But the most important aspect is that it helps with knowledge transfer and training; Documentation is a valuable resource for onboarding new team members or training existing ones. It ensures that knowledge about the decision model is not confined to individual team members.

Getting feedback should be a "recursive process", especially when your objective is to "generate actionable insights". It is also going to help you hone your other skills such as presentation, understanding of your business, and most importantly "reduce the length of time that it takes to deliver an acceptable solution."